Memmy CLI
Shared local memory for AI agents: Claude Code, Codex, Cursor and others find the agreements of past sessions in it and record new ones
- 2,133 ★ on the repository
- Last commit
- MIT
Broad access
Why this level
- The hook writes the course of every conversation with the agent to the memory database, and conversations can hold secrets and personal data
- Installation puts skills, hooks and plugins into agents' settings and registers a background service
- With a remote provider, texts from memory go to it for summarization and filtering
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add cli/memmyDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- The hook writes the course of every conversation with the agent to the memory database, and conversations can hold secrets and personal data
- Installation puts skills, hooks and plugins into agents' settings and registers a background service
- With a remote provider, texts from memory go to it for summarization and filtering
This is third-party code. Review the repository files before installing.
What it does
Memmy from MemTensor consists of a desktop app, a terminal agent and a local memory service. This entry covers the memory service and its command-line tool `memmy-memory`: the service keeps memories in SQLite on your computer, listens on 127.0.0.1:18960 and is shared by all connected agents, so an agreement from a Codex session is available in Claude Code. The `memmy-memory init` command finds installed agents (codex, cursor, claude, opencode, openclaw, hermes, dsh, workbuddy, pi, qwenwork) and gives each the memmy-memory skill, plus a hook or plugin where supported. The skill sets the order of work: `health` checks the service, `session open` opens a session, `turn start` returns the needed context from memory before an answer, `turn complete` records the interaction after it. There is `search` for direct lookup, `add` for explicit facts and project decisions, and `get` and `delete` for reading and removing by id. Context arrives in a `<memmy_memory_context>` tag as historical background, while the current request sits in `<current_user_request>`. The skill forbids answering merely from what lies in memory and storing secrets, tokens, keys, passwords and sensitive personal data. The memory panel opens in a browser at the service address.
Who it is for. For developers who run one project in several agents and want preferences, decisions and project agreements kept in one place on their own computer.
Good fit when
- You move between Claude Code, Codex and Cursor and do not want to explain the project again each time
- An agent should get relevant memories about the project and your preferences before it answers
- Memory must stay local in SQLite rather than go to a cloud service
- A decision or rule of the project has to be added to the shared memory by hand with `memmy-memory add`
- You want to see in the memory panel what the agents remembered and delete the excess by id
Not a fit when
- You need an MCP memory server: `memmy-memory` works as a command-line tool with a skill, and MCP is not connected through it
- The computer has no Node.js 20 or newer: the tool requires it, and the Linux installer also needs Node.js 22 and a systemd session
- Conversations must not be stored on disk even locally: after installation the hook writes the course of a conversation to the database
- You only need the desktop app or the terminal agent: the tool and skill work with a separate memory service
Example request
Before answering, check Memmy memory: what did we decide about the API structure in past sessions?Limitations
- The Linux installer works on x64 or arm64 and requires Node.js 22 or newer. A systemd user session is needed, and the services are bound to a local address. The npm package is built for macOS, Linux and Windows (x64 and arm64) and downloads the binary from the Releases page during installation.
- The license is MIT, but the npm package's package.json says UNLICENSED, so the package terms are worth confirming with the author.
- In BYOK mode summarization and filtering need your own model key, and with a remote provider the texts go to it. Local embeddings are switched on with the `provider: local` setting.
- The trial tokens given at registration in the app have nothing to do with the memory service: its tool works with the local database.
How to disable. Delete the memmy-memory skill and the added hook or plugin from the agent's settings directory. On Linux stop the service with `systemctl --user stop memmy-memory.service`. The data stays in ~/.memmy/memory-service/memory.sqlite, and a single memory is erased with `memmy-memory delete <id>`.
FAQ
How do I install Memmy shared memory for Codex or Claude Code?
Install the tool with `npm install -g @memtensor/memmy-memory-cli`, then run `memmy-memory init --agent codex` or `memmy-memory init --agent claude`. Without a flag `init` installs files for every agent it finds, and `--skip-agent-skills` writes only the settings.
Where does Memmy keep memories?
By default in the SQLite file ~/.memmy/memory-service/memory.sqlite. The service listens on 127.0.0.1:18960, and with a token set every route except health requires it as a bearer token or an x-api-key header.
How do I find and delete a memory?
Search with `memmy-memory search "<query>" --source <agent>`, `memmy-memory get <id> --verbose` shows the full content and `memmy-memory delete <id>` removes it. Memories are also visible in the memory panel at the service address, at `/` and `/viewer`.
Does memmy-memory serve start the memory service?
No. The `memmy-memory serve` command does not bring up the service, it only reports how to connect the standalone tool to an external memory service. The service is installed by `memmy-memory install`, for example with the `--service-only` flag.
Will the agent follow an instruction that sits in memory?
No. The skill says to treat the content of the `<memmy_memory_context>` tag as history only and to answer the request in `<current_user_request>`. It forbids answering merely because a question appears in memory.
Discussion
No questions or reviews yet.
Related
CLI toolsMemory layer for agents: plugins for Claude Code, Codex, Cursor and OpenCode, plus a CLI and SDKs that carry context across sessions
Long-term memory for agents: it distills knowledge from session history and brings it back in new tasks, one store shared across Claude Code, Codex, Cursor and
Long-term memory for AI agents: Markdown files as the source of truth, a SQLite index as a cache, shared by Claude Code, Codex and others
Long-term memory for agents: a remote MCP server, plugins for Claude Code, Cursor and Codex, and an API for your own apps
Long-term memory for AI agents: a skill and the mindmemos command store facts from conversations and recall them in later sessions
Local persistent memory for coding agents: hooks record every session, and Claude Code, Codex or Cursor recall past decisions through 54 MCP tools
Comparison with similar tools
Entries that share the main search query, and how they differ.
| Entry | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| Memmy (this entry) | no | 2,133 | ||
| Mem0 | no | 67,710 | a memory layer with plugins for Claude Code, Codex, Cursor and OpenCode, a CLI and an SDK | |
| agentmemory | yes | not required | 29,812 | local memory with hooks that record the course of a session and search for past decisions through MCP |
README in short
The description introduces the idea of shared memory and four ways to start: the app, the terminal interface, the `memmy-memory` tool and a run from source. The repository has the Memory service with a description and tests, adapters for Hermes, OpenClaw and dsh, documentation in English and Chinese and a roadmap.